Equipment Reliability: From Reactive to Prescriptive

Equipment reliability is the probability that a machine performs its intended function, without failure, for a required period under stated operating conditions. An equipment reliability strategy is the deliberate mix of maintenance approaches — from run-to-failure through prescriptive — that a plant chooses, asset by asset, to buy the most uptime per maintenance dollar.

Most plants did not choose their current strategy. They inherited it: a PM calendar someone built years ago, a maintenance crew that spends its days firefighting, and a vague ambition to "get into predictive" someday. This guide lays out the full maturity ladder — reactive, preventive, condition-based, predictive, prescriptive — with the honest costs and prerequisites of each rung, so you can decide deliberately where each asset should sit.

What is an equipment reliability strategy?

An equipment reliability strategy is a documented decision about how each class of asset will be maintained — which machines run to failure, which get scheduled preventive work, which get condition monitoring, and which justify predictive investment — based on the cost and consequence of that asset failing. It is a portfolio, not a single method.

That last point is the one most often missed. The maturity ladder below is not a race where every asset must reach the top rung. A $900 exhaust fan with a spare on the shelf belongs at the bottom of the ladder, deliberately. The bottleneck machine that stops the whole plant when it trips belongs as high as your data allows. Strategy means matching the rung to the asset — criticality, failure history, and cost of downtime decide, not enthusiasm for technology.

The inputs to that decision are unglamorous: an asset register, downtime and failure history by machine (this is where downtime tracking with reason codes pays off), and a per-minute cost of stoppage for your critical lines. If you cannot yet rank your assets by minutes lost and dollars lost, start there — the ladder can wait a month; the data cannot.

The reliability maturity ladder

Five rungs, each buying more warning time before failure — and each demanding more data, discipline, and money than the one below it.

1. Reactive (run-to-failure). Fix it when it breaks. What it costs: the highest cost per failure of any approach — collateral damage, rush-shipped parts, overtime, and unplanned downtime at the worst possible moment. Failures set the schedule, not you. Prerequisites: none, which is why it is the default. When it is right: genuinely non-critical, cheap, redundant, or easily swapped assets, chosen deliberately. Run-to-failure as a decision is a strategy; run-to-failure as a habit is just expensive.
2. Planned / preventive (PM). Service on a fixed interval — calendar, runtime hours, or cycles — regardless of condition. What it costs: planned labor and parts, plus a known inefficiency: some machines get serviced that didn't need it, and some fail anyway between services. Intervals set by usage beat intervals set by the calendar. Prerequisites: an asset register, PM task lists, and the scheduling discipline to actually execute — usually a CMMS once you pass a few dozen assets. PM completion rates are where most programs quietly die. When it is right: the workhorse approach for most assets with age- or wear-related failure modes.
3. Condition-based (CBM). Maintain when measured condition says so — vibration, temperature, pressure differential, current draw, oil condition — instead of when the calendar says so. What it costs: sensors or inspection routes, defined alarm thresholds per asset, and someone who owns responding to alerts. Unowned alerts are the classic CBM failure mode. Prerequisites: instrumentation on the asset, a way to collect readings (manual routes count), and thresholds grounded in the machine's actual baseline. When it is right: assets whose failures announce themselves through a measurable signal — most rotating equipment, for a start.
4. Predictive (PdM). Model the trend in condition data to estimate when failure will arrive, so work is scheduled before the alarm, in a planned window. What it costs: everything CBM costs, plus enough clean failure and condition history to model against — months to years of it — and analytical capability, whether an engineer, a service, or software. Cheap sensors are not the constraint anymore; trustworthy labeled history is. Prerequisites: mature CBM data collection, accurate work-order history, and patience — a predictive model trained on messy data predicts messes. When it is right: critical, instrumented assets where a surprise failure is expensive enough to justify the modeling effort.
5. Prescriptive (RxM). The system doesn't just predict the failure — it recommends or initiates the response: which action to take, when to slot it against the production schedule, what parts to stage, and it can draft the work order for a human to approve. What it costs: everything predictive costs, plus integration — the reliability system has to see the production schedule, inventory, and maintenance system to prescribe anything useful. Prerequisites: connected systems and trustworthy data end to end; prescriptions computed from bad inventory or stale schedules are confidently wrong. When it is right: as the direction of travel once prediction is working — the value is closing the gap between knowing and doing.

What the data says about each stage

The most-cited public numbers on maintenance economics come from the U.S. Department of Energy's Federal Energy Management Program O&M Best Practices Guide:

• Typical facilities still spend 40–60% of maintenance effort on reactive work; best-in-class operations hold it under 10%.
• A functioning preventive maintenance program is estimated to save 12–18% versus running reactively.
• A functioning predictive program is estimated to save a further 8–12% over preventive maintenance alone — and facilities heavily reliant on reactive work can see total savings opportunities of 30–40%.

Read those numbers with the grain of salt they deserve — they are estimates across many facility types, not a promise about your plant. But the shape of the curve is well established: each rung up the ladder cuts the cost of failures faster than it adds cost of prevention, provided the rung below it is actually working. A predictive program bolted onto a plant that skips its PMs is decoration.

How often should preventive maintenance be done?

There is no universal preventive maintenance interval — the right frequency for each task comes from the manufacturer's recommendation, adjusted by that machine's actual failure history, duty cycle, environment, and criticality. Monthly is a common default, and as defaults go it is a reasonable starting point for moderately critical equipment. But a calendar-monthly PM program has a predictable flaw in both directions: it over-services machines that run two shifts a week and under-services identical machines running around the clock.

A better sequence:

• Start from the manual. OEM intervals are the baseline — conservative, but grounded in the machine's design.
• Switch time-based tasks to usage-based where you can. Runtime hours or cycle counts track wear; calendar months do not. If you capture machine run time (even from a simple hour meter), use it.
• Let failure history move the interval. If a component keeps failing between PMs, shorten the interval or fix the underlying cause. If years of PMs find nothing to adjust, lengthen it and spend the hours on a machine that needs them.
• Weight by criticality. The bottleneck line earns tighter intervals than the machine with a warm spare, even if the equipment is identical.

The honest answer to "is monthly enough?" is: for some assets it is too much, for others dangerously little, and your own downtime and work-order history is the only thing that can tell you which. That is one more reason the data foundation comes before the strategy.

How do you measure equipment reliability?

Three metrics carry most of the weight, and they are all computable from a decent downtime log:

The MTBF/MTTR split matters because it routes the work. Frequent short failures usually call for process fixes, operator care, and better materials. Rare long failures call for spares strategy, troubleshooting aids, and captured know-how — the repair that takes four hours because only one technician knows the machine is a tribal knowledge problem wearing a maintenance costume.

How to move up the ladder without ripping anything out

The path up the ladder is incremental, and it starts with data rather than sensors:

• First, see your failures. Track every stop with timestamps and reason codes — the method in our machine downtime guide. Two honest months of this tells you your real reactive-versus-planned split and ranks your worst assets.
• Then, make PM stick. Get PM schedules, task lists, and completion tracking into one system and treat PM completion rate as a managed number. This rung is discipline, not technology.
• Bring operators into the front line of reliability. Operators notice the new vibration days before it becomes an alarm. Formalizing that — cleaning, inspection, basic care, and a fast way to flag anomalies — is autonomous maintenance, the operator-owned pillar of total productive maintenance, and it is the cheapest condition monitoring you will ever deploy.
• Instrument the critical few. Add condition monitoring where the downtime Pareto says it will pay, not everywhere. Connect the signals you already have — most plants' PLCs already know things nobody is recording.
• Only then, model and prescribe. With clean history and connected data, prediction becomes feasible and prescription becomes useful.

The connective tissue across every rung is the same: machine signals, maintenance records, downtime reasons, and production context in one place instead of five. That is the layer Harmony provides — it connects PLCs, sensors, and existing systems into one operational data layer, surfaces root-cause patterns across downtime and quality, and can act on what it finds: flag the anomaly, notify the right person, draft the work order with a human approving the action. It layers onto the ERP, CMMS, and machines you already run — no rip-and-replace. See how the platform works.

Wherever your plant sits on the ladder today, the next rung is bought with the same currency: honest data about how your equipment actually fails. Start collecting it this month.